AI Agent Hub
Back to plugins
🧠

dsh-memoryos

Memory Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install tianhao8687/dsh-memoryos

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install tianhao8687/dsh-memoryos in your DeepSeek Harness terminal to install the plugin, which is hosted at https://github.com/tianhao8687/dsh-memoryos .

About this plugin

Developers working with DSH agents often hit a wall: sessions restart, project decisions vanish, and cross-session memory—when it exists at all—is hard to benchmark against a clean baseline. dsh-memoryos bridges a local MemoryOS service into DSH so the Agent gains retrievable, updatable, time-bounded project-level long-term memory, while SQLite, the retriever, and the context compiler stay entirely inside the MemoryOS process. The Agent side sees only a slim set of HTTP-backed tools.

The plugin ships several context-delivery modes—full, progressive, compact, delta—and an optional cross-session write profile where every memory entry is anchored to a stable semantic key and a conversation excerpt, with supersede, keep_both, or reject for fact updates. Just as important is the strict no_memory arm: the model-visible MemoryOS surface drops to zero while Provider-exact input, output, and cache usage are still collected independently, giving you reproducible A/B evidence instead of vibes. Toggle memory with plain chat phrases like "开启 OS" or "关闭 OS"; the choice persists across DSH restarts with no shortcuts or shell commands required.

Built for DSH developers who need persistent project context, rigorous memory baselines for evaluation campaigns, or clean separation between provider-level token accounting and estimated memory-component attribution. Read-only by default; write access requires an explicit profile change and a reviewed policy, so everyday coding sessions stay predictable.

Use Cases

  • Recalling project decisions and technical choices after an Agent restart
  • Running clean no_memory baseline arms for memory A/B evaluation campaigns
  • Tracking provider-exact input, output, and cache tokens separately from estimated memory-component attribution

Best For

  • Developers building project-level agents on DSH
  • Researchers and evaluators who need reproducible, falsifiable memory comparisons
  • Teams that require precise token-budget control and scoped memory visibility